unslothai/unsloth · error · RecipeDatasetPublishError

Execution artifact path is not a dataset folder.

Error message

Execution artifact path is not a dataset folder.

What it means

RecipeDatasetPublishError raised when the artifact path exists but is not a directory (resolved.is_dir() is False). Publishing to Hugging Face requires a dataset folder (with metadata/config files); a plain file, a broken symlink that still resolves, or a special file at the path fails this check.

Source

Thrown at studio/backend/core/data_recipe/huggingface.py:39

    """Raised when a recipe dataset cannot be published to Hugging Face."""


def _resolve_recipe_artifact_path(artifact_path: str) -> Path:
    root = recipe_datasets_root().expanduser().resolve()
    candidate = resolve_dataset_path(artifact_path).expanduser()
    resolved = candidate.resolve(strict = False)

    try:
        resolved.relative_to(root)
    except ValueError as exc:
        raise RecipeDatasetPublishError(
            "This execution artifact is outside the Recipe Studio dataset storage."
        ) from exc

    if not resolved.exists():
        raise RecipeDatasetPublishError("Execution artifacts are no longer available.")
    if not resolved.is_dir():
        raise RecipeDatasetPublishError("Execution artifact path is not a dataset folder.")

    return resolved


def publish_recipe_dataset(
    *,
    artifact_path: str,
    repo_id: str,
    description: str,
    hf_token: str | None = None,
    private: bool = False,
) -> str:
    dataset_path = _resolve_recipe_artifact_path(artifact_path)

    try:
        from data_designer.engine.storage.artifact_storage import (
            FINAL_DATASET_FOLDER_NAME,
            METADATA_FILENAME,

View on GitHub (pinned to 203007d190)

Solutions

1. Pass the run's output directory (the artifact_path from the execution result), not a file inside it.
2. Verify with os.path.isdir(...) before calling publish.
3. If the directory was replaced, re-run the recipe.

Example fix

# before
publish_recipe_dataset(artifact_path=str(run_dir / "data.jsonl"), ...)

# after
publish_recipe_dataset(artifact_path=str(run_dir), ...)

import os
assert os.path.isdir(os.path.expanduser(artifact_path)), 'artifact_path must be the run output directory, not a file'

def is_dataset_dir(path: str) -> bool:
    p = Path(path).expanduser()
    return p.is_dir()

try:
    publish_recipe_dataset(artifact_path=p, repo_id=r)
except RecipeDatasetPublishError as e:
    if 'not a dataset folder' in str(e):
        p = str(Path(p).parent); retry(p)

When it happens

Trigger: Passing a path to a single output file (e.g. a .jsonl or .parquet file) instead of the run's output directory; the artifact dir replaced by a file of the same name.

Common situations: Clients deriving artifact_path from a reported file location rather than the run's folder; partially cleaned storage where only a marker file remains.

Related errors

• This execution artifact is outside the Recipe Studio dataset
• Execution artifacts are no longer available.
• NeMo Data Designer Hugging Face integration is not installed
• Add a Provider connection block before running this recipe.
• The Hugging Face cache folder is invalid.

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/c9f03c05aa89bf7a. Report an issue: GitHub.